Exchange Rate Movements and Employment Growth : An OCA Assessment of the CEE Economies

Exchange Rate Movements and Employment Growth: An OCA Assessment of the CEE Economies

ANSGAR BELKE^ww

Department of Economics, University of Hohenheim, Museumsflu¨gel, D-70599 Stuttgart, Germany

E-mail: belke@uni-hohenheim.de

LEO KAAS

Department of Economics, University of Vienna, A-1010 Vienna, Austria E-mail: leo.kaas@univie.ac.at

Abstract. According to the traditional ‘‘optimum currency area’’ approach, not much will be lost from a very hard peg to a currency union if there has been little reason for variations in the exchange rate in the past. This paper takes a different approach and highlights the fact that high exchange rate volatility may also signal high costs for labor markets. The impact of exchange rate volatility on labor markets in the CEECs is analyzed, revealing that volatility vis-a-vis the euro significantly lowers employment growth. Hence, eliminating exchange rate volatility could be considered a substitute for removing employment protection legislation.

Keywords: Central and Eastern Europe, currency union, euroization, exchange rate vari- ability, job creation

JEL codes: E42, F36, F42

I. Introduction

The transition process from a centrally planned economy to a market economy in Central and Eastern Europe has been accompanied by a large drop in employment. While relative improvements have been recorded in some countries for the last 2 years, the reduction in unemployment has still been modest in relation to expectations. At the beginning of the transition process it was widely assumed that the sharp, immediate increase in open unemployment would be of a temporary nature only. Most analysts expected that unemployment would soon stop rising and that on economic recovery, unemployment would level off at a relatively low level (Nesporova, 2002).

w We would like to thank Jarko Fidrmuc for supplying valuable monthly data on exchange rates, consumer price indices and annual employment data and for calculating an extensive number of trade weights. We also profited very much from comments by two anonymous referees.

ww Author for correspondence.

247

Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-284977

However, employment performance did not improve a great deal in most Central and Eastern European Countries (CEECs in the following), although this was partly due to unfavorable developments in world markets. Besides, longer-term effects of structural changes in the candidate economies have also played an important role. The countries with the largest expected in- creases in unemployment – Bulgaria, Poland and Lithuania – were among those with the highest levels. The situation in the individual countries is, of course, highly differentiated, with Hungary and Estonia at the lower end and Slovakia, Poland and Bulgaria at the upper end, with unemployment rates exceeding 15%. Yet, in all candidate countries, labor markets suffer from structural rigidities that, in combination with continued restructuring, will put a lower limit on reductions in the unemployment rates.

This paper investigates to what extent high exchange rate variability can be made responsible for the negative developments in CEEC labor markets.

Previous studies have shown that intra-European exchange rate variability has lowered the volume of trade (see, above all, the impact of exchange rate variability on trade (described in early studies by De Grauwe, 1987; Sapir and Sekkat, 1990), and has increased unemployment and reduced employ- ment, a finding that had an important bearing on the evaluation of the costs and benefits of the EMU (see, for example, Belke and Gros, 2001). More recently, Belke and Gros (2002b) have shown in the context of a project for the European Commission that exchange rate variability might also have significant negative effects on labor markets at a global level. Their results indicate that transatlantic exchange rate variability does have a significant negative impact on labor markets in the EU, and possibly also in the U.S.

The authors argue that volatility matters because employment and invest- ment decisions are characterized by some degree of irreversibility in the presence of structural rigidities. Such decisions tend to be discouraged by exchange rate variability, as can be shown in a variety of economic models. A third category of studies relates to the emerging markets as in Belke and Gros (2002a) and their investigation of the Mercosur area.

If similar results can be found for the currencies of the Central and Eastern European EU applicant countries, they would warrant a new look at the costs and benefits of joining the EMU or of using early euroization¹as a strategy to fulfill the Maastricht criterion of exchange rate stability. The main purpose of this paper is thus to provide a sound basis for an (indirect) evaluation of the costs of the present exchange rate relations of CEEC cur- rencies vis-a-vis the euro and of the benefits of individual time-paths of exchange rate policies for selected CEECs on their way towards full mem- bership of the EMU. It is argued that early entry strategies might be moti- vated with an eye to the benefits resulting from suppressed exchange rate volatility. However, policy conclusions appear to be too strong if they are based on the analysis presented in this paper. In particular, one should

additionally analyze the costs of adopting the euro caused by the loss of monetary policy autonomy. Hence, the results of this paper have to be embodied in the wide array of results gained by other current OCA analyses which show that the costs of transferring monetary policy to the ECB are likely to be low at least for some CEECs (see, for example, De Grauwe, 2003;

Fidrmuc and Korhonen, forthcoming).

The conventional view of EMU enlargement is to converge first, and durably, and then join. For the eight CEECs scheduled to join the EU in May 2004, the time frame for EMU enlargement is, thus, quite clear: EU admission also formally implies membership of the EMU. Initially, however, the new EU members will have a right of derogation concerning the introduction of the euro. When can and should derogation be lifted, i.e. when should the euro be introduced in these countries? And how can it be ensured that the transition to the euro is smooth? The earliest possible date of entry into the eurozone is 2006, if the EU’s new member states spend the prescribed 2 years in the ERM2 system immediately after EU accession.²A number of the acceding countries (e.g. the Baltic States and Slovenia) have indeed expressed a willingness to proceed to the eurozone as quickly as possible. Other countries such as the Czech Republic, Hungary and Poland are more skeptical about adopting the euro too rapidly.

Let us now outline the development of CEEC trade integration. In gen- eral, the CEECs are small, open economies. In most CEECs, external trade (imports and exports) accounts for above one third of GDP, in countries such as Slovenia (67%) or Estonia (58%) the degree of openness even exceeds 50% of GDP. Only Bulgaria (23%), Romania (15%) and, due to its larger size, Poland (26%) have a somewhat smaller openness index (Buiter and Grafe, 2002). Boreiko (2002) demonstrates the importance of trade with EMU countries for the CEECs, relating imports and exports to the eurozone to total imports and exports in 1993–2000. His tables clearly show that most of the CEECs have already reached a high share of trade with the eurozone.

In some cases – such as Hungary (0.70), Poland (0.67), Slovenia (0.67), and the Czech Republic (0.66) – the shares are close to the average of EMU intra- trade (around 0.67 in 1999–2000). The other candidate countries have lower figures (Romania: 0.63, Estonia: 0.59, Slovakia: 0.54, Latvia: 0.52, Bulgaria:

0.50, Lithuania: 0.46). These differences in openness should be kept in mind for the empirical analysis since they should, of course, influence the impact of euro exchange rate variability on the labor markets in the various candidate country. The same is true for the average degree of openness of the CEECs and the results expected from a pooled regression analysis.

The rest of the paper proceeds as follows: In Section II we derive a theoretical model to illustrate the negative relationship between exchange rate volatility and labor market performance. Section III defines our measure of exchange rate variability. Section IV presents and comments on the regression results. Section V concludes with a discussion of the implications

of our results for the design of future CEEC monetary relations with the euro zone.

II. Modeling the Impact of Exchange Rate Volatility on Labor Markets 1. MOTIVATION

Most economists would probably assume that exchange rate variability couldn’t have a significant impact on labor markets, given that the link between exchange rate variability and the volume of trade is known to be weak. However, we would argue that there are some qualifications to such a conclusion: in developing countries the level and variability of the exchange rate may be more important than in developed countries. There are several reasons why exchange rate volatility could have a strong negative impact on emerging economies and, hence, may constitute the basis for the fear of large exchange rate swings (Calvo and Reinhart, 2000).

First, the pattern of trade invoicing is different in emerging markets compared to industrial countries. Following McKinnon (1999), primary commodities are primarily dollar invoiced. Since the emerging market economies exports generally have a high primary commodity content, ex- change rate volatility should have a significant impact on the foreign trade of these countries. Additionally, capital markets in emerging markets are of an incomplete nature.³If futures markets are either illiquid or even non-existent, tools for hedging the exchange rate risk are simply not available in these countries. Another reason why emerging markets are, on average, more intolerant to large exchange rate fluctuations is due to the higher openness of these countries. When imports make up a large share of the domestic con- sumption basket, the pass-through from exchange rate swings to inflation is much higher (Calvo and Reinhart, 2000, p. 18 ff.).

In the following section we develop a full-fledged model to illustrate a mechanism that explains a negative relationship between exchange rate uncertainty and job creation.⁴ This model was originally based on the idea that uncertainty of future earnings raises the ‘‘option value of waiting’’ with decisions that concern investment projectsin general (Dixit, 1989; Belke and Gros, 2001). The model does not pretend to be close to reality. Instead, it is designed to convey the basic idea in a simple way. Moreover, our intention is to present a model that allows us to ask whether even atemporary,short-run increase in uncertainty can have a strong and lasting impact on employment, and to what extent this impact depends on labor market parameters.

2. SCENARIO A: MODELING WITH BINDING CONTRACTS

Consider a set-up in which there are three periods and asingle firmactive in an export-oriented industry decides about job creation. During the first two

periods (called 0 and 1) the firm can open a job, hire a worker and produce output that is sold in a foreign market during the following periods. If the job is created during period 0, the worker is hired for two periods (0 and 1) to produce output to be sold in periods 1 and 2. If the job is created in period 1, the worker is hired only for period 1 and output is sold in period 2.

To create a job, the firm pays a start-up costc, which reflects the cost of hiring, training and the provision of job-specific capital. After a job is cre- ated, a worker is hired and is paid a wagewabove the worker’s fallback (or reservation) wage wduring every period of employment. Alongside disutility of work, the fallback wage measures all opportunity income that the worker has to give up by accepting the job. In particular, it includes unemployment benefits, but it might also be positively related to a collective wage set by a trade union or to a minimum wage, both of which should improve the worker’s fallback position. In general, we would argue that the fallback wage should be higher in countries that are characterized by generous unemployment benefit systems, strong trade unions or minimum wage legis- lation.

In every period in which the worker is employed, he produces output to be sold in the following period in a foreign market at domestic pricepwhich has a certain componentp(the foreign price) plus a stochastic componente(the exchange rate). We assume that the foreign price is fixed (‘‘pricing to mar- ket’’), and that the exchange rate follows a random walk. In period 1, the exchange rate e₁is uniformly distributed betweenr₁andr₁. The exchange rate in period 2, e₂is uniformly distributed betweene₁r₂and e₁þr₂. An increase in r_i means an increase in uncertainty, or an increase in the mean preserving spread in period i ¼ 1, 2 (ri is proportional to the standard deviation ofei). Uncertainty can be temporary (e.g. ifr₁ > 0 andr₂ ¼ 0) or persistent (if r₂ > 0 as well). As will soon become apparent, however, the variability of the exchange rate during the second period has no influence on the result.

The wage rate w for the job is determined by the (generalized) Nash bargaining solution that maximizes a weighted product of the worker’s and the firm’s expected net return from the job. We assume that both the firm and the worker are risk-neutral. This assumption implies that risk-sharing issues are of no importance for our analysis. Thus we may assume realistically (but without loss of generality) that the worker and the firm bargain over a fixed wage ratew(which is independent of realizations of the exchange rate) when the worker is hired so that the firm bears all the exchange rate risk. A wage contract which shifts some exchange rate risk to the worker would leave the (unconditional) expected net returns unaffected and therefore has no effect on the job creation decision. Of course, if the firm were risk-averse, the assumption that the firm bears all exchange rate risk would make a post- ponement of job creation in the presence of uncertainty even more likely.

Consider first the wage bargaining problem for a job created inperiod0, in which case the worker is hired for two periods. After the job is created (and the job creation cost is sunk), the (unconditional)expectednet return of this job is equal to E₀ðS₀Þ ¼ 2p2w¼2p where p ¼ pw denotes the ex- pected return of a filled job per period (we abstract from discounting).

Denoting the bargaining power of the worker by 0 < b < 1, the firm’s net return from the job created in period 0 is⁵

E₀ðP₀Þ ¼ ð1bÞE₀ðS₀Þ c¼2ð1bÞpc: ð1Þ In order to make the problem non-trivial, the expected return from job creation in period 0 must be positive, i.e. we assume that 2ð1bÞpc>0.

Implicit in our model is the assumption that the firm and the worker sign a binding employment contract for two periods(0 and 1). Hence, job termination is not an option in case the exchange rate turns out to be unfavorable. In period 1 (after realization of the exchange rate) the conditional expected surplus from job continuation is E₁ðS₁Þ ¼ pþe₁ which may be negative if the exchange rate falls below p<0.

If the firm waits untilperiod 1, it retains the option of whether or not to open a job. It will create a job only if the exchange rate realized during period 1 (and so expected for period 2) is above a certain threshold level, or barrier, denoted by b. Given that employment in period 1 yields a return in period 2 only, this profitability barrier is defined by the condition that the (condi- tional) expected net return to the firm is zero:

ð1bÞðpþbwÞ c ¼0 or b ¼c=ð1bÞ þwp ¼c=ð1bÞ p:

ð2Þ Whenevere₁ b, the firm creates a job in period 1, and the conditional expected net return to the firm is E₁ðP1Þð1bÞðpþe₁Þ c0. Whenever e₁ < b, the firm does not create a job in period 1, and its return is zero.

Hence, whenever both events occur with positive probabilities (i.e. whenever r₁ > b > r₁),⁶the unconditional expected return of waiting in period 0 is given by

E₀ðP1Þ ¼ ½ðr1þbÞ=ð2r1Þ0þ ½ðr1bÞ=ð2r1Þ½ð1bÞðpþ ðr1þbÞ=2Þ c;

ð3Þ where the first element is the probability that it will not be worthwhile opening a job (in this case the return is zero). The second term represents the product of the probability that it will be worthwhile opening the job (because the exchange rate is above the barrier) and the average expected value of the net return to the firm under this outcome. Given condition (2) this can be rewritten as

E₀ðP1Þ ¼ ð1bÞðr1bÞ²=ð4r1Þ: ð4Þ This is the key result since it implies that an increase in uncertainty in- creasesthe value of waiting, given that Equation (4) is an increasing function

of r₁.⁷ Asr₁ increases, it becomes more likely that it is worthwhile waiting until more information is available about the expected return during period 2. The option not to open the job becomes more valuable with more uncertainty. The higher the variance, the higher the potential losses the firm can avoid and the higher the potential for a very favorable realization of the exchange rate, with consequently very high profits. It is clear from (1) and (4) that the firm prefers to wait if and only if

ð1bÞðr1bÞ²=ð4r1Þ>2ð1bÞpc: ð5Þ As the left hand side increases in r₁, the firm delays job creation if ex- change rate uncertainty is large enough. The critical value at which (5) is satisfied with equality can be solved as⁸

r₁¼3pc=ð1bÞ þ2 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pð2pc=ð1bÞÞ

p : ð6Þ

Whenever r₁ > r₁, firms decide to postpone job creation in period 0.

Since r₁ is increasing inp (and thereby decreasing in the fallback wage w), decreasing in the cost of job creation c and decreasing in the worker’s bar- gaining powerb, we conclude that the strong position of workers in the wage bargain (reflected in a high fallback wage or in the bargaining power parameter) and higher costs of hiring increase the option value of waiting and make a postponement of job creation more likely. Thus,the adverse impact of exchange rate uncertainty on job creation and employment should be stronger if the labor market is characterized by generous unemployment benefit systems, powerful trade unions, minimum wage restrictions or high hiring costs. The adverse employment effects of these features have been confirmed empirically in various studies, and there are many other theoretical mechanisms to ex- plain them (see, for example, Nickell, 1997). What our simple model shows is that these features also reinforce the negative employment effects of con- temporaneous and short spikes of exchange rate uncertainty. In sum, we retain two conclusions from the model. First, even a temporary spike in ex- change rate variability can induce firms to wait before creating jobs (for precisely this reason, the level of the exchange rate loses explanatory power at the same time). Second, the relationship between exchange rate variability and employment should be particularly strong if the labor market is char- acterized by rigidities that improve the bargaining position of workers. A stronger fallback position of workers increases the contract wage, lowers the net returns to firms and induces firms to delay job creation in the face of uncertainty.

3. SCENARIO B: MODELING WITHOUT BINDING CONTRACTS

Our argument in Section II.2 rests on the assumption that workers cannot be fired immediately if the exchange rate turns out to be unfavorable. Hence, sunk wage payments are associated with the decision to hire a worker. These

sunk costs and, consequently, the impact of uncertainty on job creation become more important if there are high firing costs. However, even if there are no firing costs and if workers can be laid off at any point in time, ex- change rate uncertainty should have a direct impact on job destruction. To illustrate this proposition, consider the scenario of a labor market in which the firm and the worker can sign a contract only for one period and keep the option to terminate the work relationship whenever it becomes unprofitable.

In period 1, the conditionally expected surplus of job continuation isp þ e₁ which is positive whenever e₁ > p. Hence, whenever uncertainty is large enough (r₁ > p), there is job destruction in period 1 with probability ðr₁pÞ=ð2r₁Þ. The (unconditional) expected net return to the firm from a job created in period zero (and with the option of destruction in period one) is therefore

E0ðP0Þ ¼ ½ð1bÞpc þ ½ðr1pÞ=2r10þ ½ðr1þpÞ=2r1ð1bÞ½pþ ðr1pÞ=2Þ;

ð7Þ where the first term is the expected return from the job in period one, whereas the second and third term represent the expected surplus from the job in period two (after destruction or after continuation in period one) under the assumption r₁ > p. If r₁ < p, the job would never be destroyed, and the expected net return is, as before, 2ð1bÞpc. Hence, after rearranging (7), the expected net return from a job created in period zero can be written

E₀ðP0Þ ¼ 2ð1bÞpc; if r₁<p, ð1bÞðpþ ðr₁þpÞ²=ð4r₁ÞÞ c; if r₁p.

ð8Þ On the other hand, if the firm waits until period 1, the (unconditional) ex- pected net return is, as in Section II.2,

E₀ðP₁Þ ¼ maxð0;ð1bÞpcÞ; ifr₁<jpc=ð1bÞj, ð1bÞðr1þpc=ð1bÞÞ²=ð4r1Þ; ifr₁ jpc=ð1bÞj.

ð9Þ It is now easy to see that the firm never delays job creation. First, if r₁ jpc=ð1bÞj<p, the firm never destroys a job in period 1, and so we have E₀ðP0Þ>E₀ðP1Þ. Second, if r₁p, the condition E₀ðP0Þ>E₀ðP1Þ means that

4r₁ðpc=ð1bÞÞ þ ðr1þpÞ²>ðr1þpc=ð1bÞÞ² ð10Þ which turns out to be equivalent toð2ð1bÞpcÞðc=ð1bÞ þ2r₁Þ>0 and which is satisfied because of our assumption 2ð1bÞpc>0. Hence, the firm does not delay job creation in this case either. Finally, if jpc=ð1bÞj<r₁ <p, the conditionE₀ðP₀Þ>E₀ðP₁Þ means that

4r₁ð2ð1bÞpcÞ ð1bÞðr1þpc=ð1bÞÞ²>0: ð11Þ But since this inequality is satisfied at the boundaries r₁¼p and r₁ ¼ jpc=ð1bÞjand since the left hand side is a concave function ofr₁,

the inequality is also satisfied in the intervaljpc=ð1bÞj<r₁<p. Hence, firms always prefer to create a job in period zero, and so exchange rate uncertainty has no impact on job creation.

However, since there is job destruction with probability ðr₁pÞ=ð2r₁Þ (wheneverr₁>p),the probability of job destruction increases in uncertainty.

Hence, there is also a negative impact of exchange rate uncertainty on employment in this case. Moreover, this effect is more pronounced if the worker’s fallback wage is higher (if p is smaller). Therefore, the basic con- clusions of our model remain valid. Thus, we can show that, perhaps against common intuition (but as an additional interesting innovation in our paper), uncertainty does not lead to delays but to spurts in the decision to fire. In contrast to the usual models of investment under uncertainty in the Dixit- Pindyck style, the impacts of increased uncertainty on employment are unambiguously negative and can be directly tested empirically.

4. DOES THE MODEL APPLY TO THE CEEC LABOR MARKETS?

According to our model, the relationship between exchange rate variability and employment should be particularly strong if the labor market is char- acterized by rigidities that, for example, improve the bargaining position of workers. The labor markets in most of the current EU members are widely considered to be rigid enough to give leeway to the functioning of the mechanism explained in the model. Where do the candidates stand in this respect?

Riboud et al. (2002) have assessed the flexibility of labor market institu- tions in six CEE candidate countries: the Czech Republic, Estonia, Hungary, Poland, Slovakia and Slovenia.⁹According to their findings, based on a large scale of indicators for regular contracts, temporary contracts and collective dismissals, these countries range somewhere in the middle of the flexibility scale compared to the OECD economies. They do not reach the levels of flexibility of the UK, Ireland and Denmark, but exhibit much greater flexi- bility than the Club Med countries, France and Germany.¹⁰As regards their unemployment insurance systems, the CEECs seem to be less generous than the OECD or the EU countries. They also spend less on both passive and active employment policies. In terms of the role of the unions in the wage negotiation process, the candidates range somewhere in the middle of the OECD countries. They do, however, have extremely high payroll and other taxes, which exceed even the highest levels in the EU. Even more important in our context is the fact that they have strong employment protection legis- lation (Table I).

These results are consistent with findings by Belke and Hebler (2002) and Cazes (2002) which state that Central European countries have adopted labor market institutions, institutional arrangements and legal frameworks that

TableI.LabormarketflexibilityintheCEECs:Howhigharethecostsofjobcreationandthefallbackwage? Employmentprotectionlegislationa Unemployment insuranceTaxes Regular empl.Temporary empl.Collective dismissalsEPL StrictnessbBenefitrepla- cementratioBenefitdura- tion(months)Payrolltax rate(%)Totaltax rate(%) CzechRepublic2.80.54.32.150647.573.4 Estonia3.11.44.12.6103–633.063.3 Hungary2.10.63.41.7641244.081.5 Poland2.213.924012–2448.280.0 Slovakia2.61.44.42.4606–1250.081.0 Sloveniac 3.4(2.9)2.4(0.6)4.8(4.9)3.5(2.3)633–2438.069.1 CEECaverage2.71.24.12.44843.474.7 EUaveraged 2.42.13.22.46023.553.0 OECDaverage2.01.72.92.05819.545.4

PassivepoliciesActivepoliciesUnions %ofGDPSpendingper unemployed%ofGDPSpendingper unemployedUniondensity (%)eUnioncover- ageindexfCoordina- tion-unionsCoordination- employers CzechRepublic0.310.040.190.0242.8211 Estonia0.080.010.080.0136.1221 Hungary0.560.060.400.0460.0312 Poland1.710.120.490.0333.8321 Slovakia0.540.050.560.0561.7322 Slovenia0.890.110.830.1160.0333 CEECaverage0.680.060.420.0449.0 EUaverage1.730.261.160.1644.4 OECDaverage1.430.230.920.1439.6 a 1:minimumprotection,6:maximumprotection. b Weightedaverageofthefirstthreecolumns. c Numbersinbracketsrefertothenewlaborcodeifapproved. d EUaveragewithoutLuxembourgandGreece. e Percentageofsalariedworkersthatbelongtoaunion. f 1:lessthan25%ofsalariedworkersarecoveredbycollectiveagreements,2:between26and69%arecovered,3:70%ormorearecovered. Source:Hobza(2002)andRiboudetal.(2002).

share many common features with present EU member countries. This trend clearly increases job creation costs. It is further supported by the fact that the CEECs have been required, prior to their entry into the EU, to align their legislation with the acquis communautaire, which includes a number of pro- visions regarding labor market regulations. This kind of legislation favors employment protection while taxing employers heavily. Hence, the trans- mission channel from exchange rate variability to labor market performance that we have described seems to be relevant in the case of the CEECs as well.

The next step is to address whether different measures of exchange rate volatility – both nominal and real effective volatility vis-a-vis the 31 most important trade partners and the bilateral volatility of the nominal and real euro exchange rate – have any ability to explain the residuals of employment regressions for CEEC economies. Up to now, the literature examining the link between exchange rate variability and labor market performance in emerging markets is rather thin. Hence, we begin by presenting and com- menting on some initial results.

III. Data and Definitions

In order to test empirically for the conjectured impact of exchange rate variability on labor-market performance, we employ a panel of 10 Central and Eastern European countries, namely Bulgaria (BG), the Czech Republic (CZ), Estonia (EE), Hungary (HU), Latvia (LV), Lithuania (LT), Poland (PL), Romania (RO), Slovakia (SK), and Slovenia (SL). We have not left out the two EU latecomers Bulgaria and Romania because Bulgaria, at least, is often said to be a clear case for euroization.

The nominal variability of the currency of each of the 10 CEECs that have applied for EU membership is measured by taking, for each year, the stan- dard deviation of the 12 month-to-month changes in the logarithm of its nominal exchange rate against the currencies of the countries’ main trade partner countries. The construction of the real variability variable follows an analogous scheme. The nominal exchange rates are deflated with the CPI.

The standard deviations based on bilateral rates are then aggregated in one composite measure of exchange rate variability (denoted by ‘‘VOL’’ below) using the weights that approximate the importance of these currencies in trade with the 31 most important trade partners for the period, 1993-2002.

The average trade weight of CEEC X with country Y (Y¼Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, the Nether- lands, Portugal, Spain, Sweden, the UK, Bulgaria, the Czech Republic, Es- tonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, Slovenia, Croatia, Belarus, Russia, Ukraine, Switzerland, the U.S., Turkey) is calcu- lated as 100 times the exports to country Y plus imports from country Y divided by total exports to the ‘‘world’’ plus total imports from the ‘‘world’’.

In our definition, the aggregate ‘‘world’’ corresponds to the sum of countries Y.¹¹We did not use the annually changing trade weights since our volatility measure would then change to the same degree as the change in trade pattern varies.

Based on the monthly CPI series for the 31 most important trade partners, the nominal bilateral exchange rates of these countries vis-a-vis the U.S.

dollar and their trade weights, we calculated the following volatilities of the exchange rate:

10 times 30 volatilities of the nominal bilateral exchange rate, 10 times 30 volatilities of the real bilateral exchange rate,

10 effective volatilities of the nominal exchange rate (weighted bilateral volatilities), and

10 effective volatilities of the real exchange rate (weighted bilateral vol- atilities).

It should be emphasized that the first two series refer to the exchange rate volatility ‘‘vis-a-visthe euro’’. This is calculated as the volatilityvis-a-vis the DM from, 1992:01 to1998:12, andvis-a`-visthe euro from, 1999:01. We prefer to aggregate the individual standard deviations instead of using a standard deviation of an average or effective exchange rate because there is extensive evidence that CEEC exporters have priced to market. When using an average exchange rate, the zloty, for example, might remain constant because depreciation against the euro would compensate for appreciation against the Bulgarian lewa. Polish firms would not necessarily be indifferent to whether a situation arises in which the average exchange rate is constant because the zloty/euro and the zloty/lewa are constant, and another in which the swings in these two bilateral rates just happen to cancel each other out.

We use monthly rather than daily exchange rates to calculate volatility to ensure consistency throughout our entire sample period. Another reason for preferring this measure to shorter-term alternatives (e.g. daily variability) was that, while the latter might be important for financial actors, they are less relevant for export or employment decisions. The drawback of monthly ex- change rates is that we had to use annual data to have a meaningful measure of variability, leaving us with only nine observations for each country.¹²

We use actual as opposed to unanticipated rates, since in order to be consistent with our model as described in Section II, we assume that the exchange rate follows a random walk. Thus, actual and unanticipated ex- change rate changes should be the same. We feel justified making this assumption since extensive research based on work by Meese and Rogoff (1983) and Meese (1990) has shown that the random walk model outperforms other standard exchange rate models in out-of-sample forecasting. This still holds even when seemingly relevant economic variables are included. Our sample covers the period, 1992–2001 in order to exploit all reliable data

information. However, in view of the financial turmoil in the early years of transition, our estimations mostly exclude the year, 1992. The average ex- change rate variability for each of the 10 CEECs under investigation is plotted in Figure 1 (percent per month). Peaks usually occur in the year 1998, with the two non-EU acceding countries Bulgaria and Romania as clear outliers with high double-digit realizations. Low volatility values typically appear at the end of the sample, especially in, 2000 and 2001. Effective real volatility has decreased for countries that used exchange rate arrangements close to fixed rates, but remained high for Poland and Romania and was

-4 -3 -2 -1 0 1

1 2 3 4 5 6 7 8

NOMEFFVOL

DEMP_AVG

BG CZ

EE HU

LT LV

PL

RO SK

SL

-4 -3 -2 -1 0 1

0 4 8 12 16 20

REALEFFVOL

DEMP_AVG

SL SK CZ

HU

PL BG

RO LV EE LT

-4 -3 -2 -1 0 1

0 1 2 3 4 5 6 7

NOM_DMEUR_VOL

DEMP_AVG

BG S L

S K CZ

HU P L

RO LV

LT E E

-4 -3 -2 -1 0 1

0 1 2 3 4 5

REAL_DMEUR_VOL

DEMP_AVG

S L CZ S K

HU

P L RO BG

LV LT

E E

Figure 1. Employment growth (in %) and exchange rate volatility (10 Central and Eastern European Countries, average, 1992–2001). (a) Effective nominal exchange rate volatility. (b) Effective real exchange rate volatility. (c) Nominal exchange rate volatility vis-a`-visthe Euro. (d) Real exchange rate volatilityvis-a-visthe Euro.

quite high for Latvia and Lithuania (for a similar observation, see Boreiko, 2002, p. 14 ff.). In the case of the countries with macroeconomic instability and high inflation, an inspection of our data reveals that variation in the bilateral real exchange rate is large and much higher than nominal exchange rate variability. This is somewhat surprising given the fact that PPP is usually more likely to hold for countries with very high rates of inflation, which would suggest that real exchange rate variability should be smaller. However, it seems obvious that high real exchange rate variability signals weak mac- roeconomic management rather than a need for the real sector to be adjusted.

In contrast to our previous studies and strictly in line with our model, we leave out unemployment and limit our empirical analysis to the impact of exchange rate variability on employment. We use the yearly average of employment in thousands.

The existence of a significant unofficial sector in the CEECs should not matter too much for regressions if one uses changes in employment. More- over, data on employment naturally refers only to official employment, i.e.

those officially declared and thus subject to social security contributions, income tax, and all official labor market regulations. This implies that we do not take into account the potentially very large black or underground economy for data availability reasons. Focusing on the official labor market is, however, entirely appropriate. In the black economy the costs of firing are presumably much lower because official employment regulations do not apply. This implies that our model of firing costs applies mainly to official employment and we would expect volatility to be mainly a deterrent to official employment. Data on (official) employment is usually much more accurate than data on unemployment because the definition of who is looking for work but unable to find it, often changes. Moreover, geo- graphical coverage of the unemployment statistics changes over time as well.

At times the national unemployment data mainly reflect data from one or two major provinces. Employment data, in contrast is usually nationwide because it encompasses all people on the social security registers.

As a cyclical control variable in the employment equations, we include the real growth rate of the gross domestic product, in percent. The development of wage costs is approximated by the real growth rate of average gross monthly wages in percent. With the exception of Estonia, Latvia and Lith- uania, where we used Eurostat and national sources, the data for employ- ment, GDP growth and wage costs are taken from the CEEC data set compiled by the Vienna Institute for International Economic Studies.

In order to convey a broad-brush view on the data set and some of the possible correlations, four scatter plots are presented in Figure 1 showing cross-plots of our measure for total economy employment growth in percent against exchange rate volatility. All variables are averaged over the period, 1993–2001.

As expressed by the simple scatter plots relating average employment growth (in percent) to the average volatility measure, there seems to be some initial evidence that the conjectured negative relationship between exchange rate volatility and employment holds in a cross-country perspective for the CEECs. What matters is that the overall relationship in the figures is downward sloping (non-vertical and non-horizontal). Hence, we fit a very preliminary bivariate regression of average employment growth (in percent) on an average of four different measures of exchange rate variability. In none of these four cases can we reject the hypothesis of a negative relationship according to the fitted regression lines. The example of Estonia shows that the introduction of a currency board does not protect an economy from the negative impact of effective exchange rate variability. The same is true with respect to Latvia with its exchange rate fixed to the SDRs.

However, in order to investigate the validity of our hypothesis more deeply, we will conduct fully specified regressions in the following section.

Our formal empirical analysis is based on tests of the non-stationarity of the levels and the first differences of the variables under consideration, i.e. total economy employment, the different operationalizations of exchange rate volatility, and the real growth rate of average gross monthly wages.¹³The test applied is the first widely used panel data unit root test by Levin and Lin (1992). The results indicate that only (the log of) employment has to be differenced once to become stationary. That is, we estimate panel regressions for employment growth.

Our unit root tests reveal evidence of a stationary behavior on the levels of exchange rate volatility and real wage growth. Hence, we use employment growth and the levels of exchange rate volatility in the following pooled estimations.

IV. Empirical Model and Results

Based on our theoretical arguments, we conjecture that, controlling for the usual key variables on the labor market, we can show in a cross-country panel analysis of Central and Eastern European countries that exchange rate variability worsens labor-market performance. To test for a significant neg- ative relationship between exchange rate variability and labor-market per- formance, we undertake a fixed effects estimation. By doing so, we account for different intercepts and, hence, different natural rates of employment estimated for each CEEC.¹⁴

In the literature, random effects models are sometimes additionally implemented, mainly because fixed effects models and country dummies are costly in terms of lost degrees of freedom. We decided to dispense with such an exercise because our sampled cross-sectional units could not be drawn from a large population. Moreover, following our main argument in Section

II, there is no reason to assume that country-specific constants in the employment equations are random a priori. Hence, random effects models will probably have to be excluded a priori for many richer specifications of the empirical model from this perspective as well.

However, for reasons of methodological correctness, we also performed Hausman specification tests to check empirically whether fixed effects is the correct estimation procedure (against the possible alternative of random ef- fects). For this purpose, we computed the Hausman test statistic for testing the null hypothesis of random effects against the alternative hypothesis of fixed effects (Hausman, 1978; Stata Corporation, 2003, p. 51 ff.). We are concerned with four specifications according to which employment growth is a function of the real GDP growth rate and one of four different measures of exchange rate variability (effective volatility of nominal and real exchange rates and the nominal and real exchange rate volatility ‘‘vis-a-vis the euro’’).

Moreover, we differentiate between static and dynamic specifications. In our context, ‘‘dynamic’’ means the additional inclusion of the growth of employment lagged two periods as an instrument for the lagged endogenous variable. The results, which reveal overwhelming evidence in favor of a fixed effects estimation, are displayed in Table II (panels A and B). Hence, we feel legitimized to dispense with random effects estimations and to focus on fixed effects models.

The empirical model we use can now be described by the usual form:¹⁵

yit¼aiþx⁰_itb_iþeit; ð12Þ

with yit as the dependent (macroeconomic labor market) variable;xit and b_i as k-vectors of non-constant regressors (e.g. exchange rate variability) and parameters for i¼1;2;. . .;Ncross-sectional units andt¼1;2;. . .;T as the periods for which each cross-section is observed. Imposing ai ¼aj ¼a, a pooled analysis with common constants is nested in this specification.

In order to test for significance of the impact of exchange rate volatility on labor-market performance in CEECs, we separate our analysis into three logical steps. We note that basing the analysis onlevelsof employment as an endogenous lagged variable is problematic for, at least, three reasons. First, employment time series might be plagued by non-stationarity problems (see Section III). Second, one has to take account of the well-known problem of endogenous lagged variables in the context of panel analyses (group effects).

This is usually achieved by taking first differences, which is a further reason why we conducted our analysis in these terms. Third, the theoretical interest is in a link between the level of exchange rate volatility and the growth of employment. According to this specification, a one-time shock in exchange rate volatility results in a permanent reduction in the employment level. This exactly mirrors the central persistence implication of the model which pos- tulates that even a short-term increase in exchange rate uncertainty leads to

less hiring and more firing. The dynamic implications of our specification are thus acceptable for temporary shocks, i.e. spikes, in exchange rate variability, which were emphasized in our model.

In principle, our panel data set need not be applied to astaticspecification (in the following tables this corresponds to the first column for each volatility measure). Especially with respect to the well-known path-dependence of employment, it is advisable to test for dynamic effects as well. In order to capture the speed of adjustment of labor markets, we use the option to include lagged employment variables in the set of regressors throughout this paper. The corresponding setting with respect to a representative regression equation for one cross-section out of the whole system (described by the index i) can be described as follows:

yit¼aiþx⁰_itb_iþdyi;t1þeit: ð13Þ However, for estimating our first-order model, substantial complications have to be taken into account. Problems arise due to the heterogeneity of the cross-sections analyzed (Greene, 2000, p. 582 ff). The main problem to be treated here is the correlation of the lagged dependent variable ‘‘level of employment’’ with the disturbance, even if the latter does not exhibit

Table II. Hausman tests for fixed vs. random effects

Measure of EXR variability used Test statistics Empirical realization Panel A: Static specifications

Effective volatility of nominal exchange rates Chi-sqr(2) 4.04

P-value 0.13

Effective volatility of real exchange rates Chi-sqr(2) 11.63

P-value 0.00

Nominal EXR volatility ‘‘vis-a`-visthe euro’’ Chi-sqr(2) 3.73

P-value 0.15

Real EXR volatility ‘‘vis-a`-visthe euro’’ Chi-sqr(2) 4.96

P-value 0.08

Panel B: Dynamic specifications

Effective volatility of nominal exchange rates Chi-sqr(2) 17.30

P-value 0.00

Effective volatility of real exchange rates Chi-sqr(2) 21.13

P-value 0.00

Nominal EXR volatility ‘‘vis-a`-visthe euro’’ Chi-sqr(2) 11.83

P-value 0.01

Real EXR volatility ‘‘vis-a`-visthe euro’’ Chi-sqr(2) 18.98

P-value 0.00

autocorrelation itself. While taking first differences allows the elimination of heterogeneity, i.e. the group effects, the problem of the correlation between the lagged dependent variable and the disturbance still remains. Moreover, a moving-average error term now appears in the specification. However, the treatment of the resulting model is a standard application of the instrumental variables approach. The transformed model looks as follows:

yityi;t1¼ ðxitxi;t1Þ⁰b_iþdðyi;t1yi;t2Þ þ ðeitei;t1Þ ð14Þ Arellano (1989) and Greene (2000), for instance, recommend using the dif- ferences ðy_i;t2y_i;t3Þ or the lagged levels y_i;t2 and y_i;t3 as instrumental variables forðyi;t1y_i;t2Þin order to derive a simple instrumental variable estimator. The remaining variables can be taken as their own instruments.

Arellano (1989) gives some theoretical and empirical support in favor of preferring levels to differences as instruments. As our second step of analysis, we therefore implement this procedure within a dynamic framework (in the following tables this corresponds to the second column for each volatility measure). Finally, as a third step, we conduct robustness tests by also including variables representing labor-market rigidities. Throughout our regressions, we take employment growth as the regressand.

We rely on feasible generalized least squares (FGLS) estimates of a model assuming the presence of cross-sectional heteroscedasticity and autocorrela- tion but without correction for contemporaneous correlation.¹⁶ One might argue that uncorrelatedness across our cross-sectional units (countries) is too strong an assumption because our model assigns the same parameter vector to all units in the common coefficients case, in which seemingly unrelated regression (SUR) estimates of a model with heteroscedasticity and cross- sectional correlation would be suitable. However, in view of the fact that correlations across countries might become relevant mainly in the case of symmetric shocks to the labor markets and that the probability of the latter might be small in our large sample (see, for example, Babetski et al., 2002), it is legitimate to apply an FGLS specification that assumes solely the presence of cross-section heteroscedasticity (Table III). In order to be consistent in the sense of accounting for the possibility of symmetric shocks (i.e. contempo- raneous correlation), we nevertheless apply the seemingly unrelated regres- sion technique in our regression analysis as well (Tables IV and V).

The structure for presenting the estimation results is the same for Tables III–V with the exact specifications of the pooled estimation equations being described in the tables themselves. Half of the specifications include a lagged endogenous labor-market variable. All specifications contain contempora- neous real GDP growth with or without its lagged value as cyclical control, different measures of exchange rate variability and the estimates of the country-specific constants.¹⁷The number of lags of the relevant variables was determined by the estimation itself. As in our previous studies, we limited

TableIII.Impactofexchangeratevariabilityonthegrowthofemployment–FGLSestimatesfor10CEECs(fixedeffects) Regressors(1)(2)(3)(4)(5)(6)(7)(8)(9)(10) Instrumentfor employmentgrowth(–1) /0.10/0.14*///0.25***/0.18* RealGDP growthrate

0.26***0.16**0.21***0.15**0.24***0.20***0.25***0.20***0.27***0.19*** Measuresofexchangeratevolatility Effectivevolatilityofnominal exchangerate)0.20** ()2))0.14** ()2))0.16* ()2) Effectivevolatilityofreal exchangerate)0.11*** ()2))0.09*** ()2))0.06* ()2) Volatilityofnationalcurrencyvis-a`-vis euro(DM)(nominalexchangerate))0.17** ()2))0.15* ()2) Volatilityofnationalcurrencyvis-a`-vis euro(DM)(realexchangerate))0.36*** ()2))0.34*** ()2) CommonAR-errorassumed··

Fixedeffects _BG0.550.040.590.130.53)0.08)0.000.440.460.94 _CZ)0.73)0.70)0.22)0.42)0.79)0.69)0.83)0.66)0.57)0.34 _EE)3.12)2.42)2.15)1.67)3.40)2.90)3.49)2.38)3.02)2.05 _HU)0.900.29)0.410.68)0.170.11)1.04)0.15)0.670.27 _LT)2.01)1.62)1.09)0.44)2.37)1.51)2.15)1.45)1.61)0.93 _LV)2.18)0.58)0.400.32)0.95)0.29)2.39)1.57)1.66)0.91 _PL)1.40)0.77)0.90)0.47)1.27)1.10)1.47)1.08)0.91)0.49 _RO0.480.450.790.800.09)0.120.250.061.190.87 _SK)0.55)1.160.13)0.64)0.74)0.42)0.66)0.69)0.42)0.33 _SL0.320.120.520.170.380.470.200.120.400.44 WeightedStatistics R2 0.320.280.320.320.400.430.310.290.370.40 F-statistics2.821.662.912.033.113.532.642.903.493.26 Wooldridge–^t(P-value)1.17 (0.25)

1.26 (0.21)

1.21 (0.23)

0.80 (0.43)

//1.35 (0.18)

1.49 (0.14)

0.59 (0.56)

1.54 (0.13) Totalpanelobservations78647964686978727872 Sample1993– 20011994– 20011993– 20011994– 20011993– 20011993– 20011993– 20011994– 20011993– 20011994– 2001 Thetermðyi;t1yi;t2Þisinstrumentedbythegrowthofemploymentlaggedtwoperiods.Numbersinbracketsrefertotheoptimallag. ***,**,*indicate1,5,10%significance.

TableIV.Impactofexchangeratevariabilityonthegrowthofemployment–SURestimatesfor10CEECs(fixedeffects) Regressors(1)(2)(3)(4)(5)(6)(7)(8)(9)(10) Instrumentforthe growthofemployment()1)

/0.06/0.09***///0.05/0.04 RealGDPgrowthrate0.19***0.11***0.20***0.11***0.06***0.12***0.17***0.10***0.16***0.09*** Measuresofexchangerate volatility Effectivevolatilityof nominalexchangerate)0.20*** ()2))0.16*** ()2))0.29*** ()2) Effectivevolatilityofreal exchangerate)0.08*** ()2))0.08*** ()2))0.13*** ()2) Volatilityofnationalcurrency vis-a`-viseuro(DM) (nominalexchangerate)

)0.16*** ()2))0.10*** ()2) Volatilityofnationalcurrency vis-a`-viseuro(DM) (realexchangerate) )0.35*** ()2))0.08* ()2) AR-errorassumed